Photonic PCFBench - Design Direction Hole: leaderboard
PCFBench v1 Appendix full results; 26 isolated tasks across six categories; temperature0.1, task-specific64–2048token budgets; native fractions/errors. Paired structure+field and Field-only contexts separate. Closed-loop success uses expert sign/magnitude constraints, not physical resimulation; Human Score mean of six dimensions/two experts on120samples. Source-disputed summary/appendix cells and unbound checkpoint variants withheld.
Metric: Accuracy. Source: arxiv.org. Saturation forecast: Not forecast. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 7B Instruct | 0.73 | #650 |
| 2 | Qwen 2.5 VL 72B Instruct | 0.66 | #383 |
| 3 | O3 | 0.65 | #132 |
| 4 | GPT-5 | 0.62 | #105 |
| 5 | Claude Opus 4.5 (Thinking) | 0.61 | #87 (Claude Opus 4.5) |
| 6 | Gemini 3 Flash | 0.59 | #103 |
| 7 | GPT-4o Mini | 0.58 | #598 |
| 8 | GPT-5 Mini | 0.55 | #193 |
| 9 | Claude Opus 4.5 (Non-reasoning) | 0.55 | #87 (Claude Opus 4.5) |
| 10 | O4 Mini | 0.49 | #185 |
| 11 | Qwen 2 VL 72B Instruct | 0.49 | #551 |
| 12 | Claude Sonnet 4.5 (Thinking) | 0.47 | #148 (Claude Sonnet 4.5) |
| 13 | Claude Sonnet 4.5 (Non-reasoning) | 0.4 | #148 (Claude Sonnet 4.5) |
| 14 | Qwen 3.5 27B | 0.31 | #192 |
| 15 | Qwen 3.5 9B | 0.28 | #391 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=photonic-pcfbench-design-direction-hole · How It Works · Data refreshed daily, snapshot 2026-11-05.